Structure, composition and dynamics of a calcareous grassland metacommunity over a 70-year interval


  • Adrian C. Newton,

    Corresponding author
    1. Centre for Conservation Ecology and Environmental Science, School of Applied Sciences, Bournemouth University, Poole BH12 5BB, UK
      Correspondence author. E-mail:
    Search for more papers by this author
  • Robin M. Walls,

    1. BSBI Vice-County Recording, 10 Old Brickfields, Broadmayne DT2 8UY, UK
    Search for more papers by this author
  • Duncan Golicher,

    1. Centre for Conservation Ecology and Environmental Science, School of Applied Sciences, Bournemouth University, Poole BH12 5BB, UK
    2. Departamento de Ecología y Sistemática Terrestre, El Colegio de la Frontera Sur, Carretera Panamericana y Periférico Sur s/n, San Cristóbal, Chiapas 29290, Mexico
    Search for more papers by this author
  • Sally A. Keith,

    1. Centre for Conservation Ecology and Environmental Science, School of Applied Sciences, Bournemouth University, Poole BH12 5BB, UK
    Search for more papers by this author
  • Anita Diaz,

    1. Centre for Conservation Ecology and Environmental Science, School of Applied Sciences, Bournemouth University, Poole BH12 5BB, UK
    Search for more papers by this author
  • James M. Bullock

    1. Centre for Ecology and Hydrology, Benson Lane, Wallingford, Oxfordshire OX10 8BB, UK
    Search for more papers by this author

Correspondence author. E-mail:


1. Calcareous grasslands are communities of high conservation value, often characterized by high plant species richness. These grasslands have experienced a major decline in area throughout Europe, principally resulting from agricultural intensification. Although they have been the focus of extensive previous research, few attempts have been made to examine the long-term dynamics of multiple communities at the landscape scale.

2. To assess long-term change in the structure and composition of a calcareous grassland metacommunity, 88 extant sites first surveyed by R. Good in the 1930s were resurveyed in 2009. Values of α-, β- and γ-diversity were compared between the two surveys, using a one-way analysis of similarity (ANOSIM) and non-metric multidimensional scaling. Elements of metacommunity structure (EMS) analysis was used to identify metacommunity structure, and changes in metacommunity composition were related to plant traits.

3. Analyses indicated that α-diversity increased over time, with mean (±SD) species richness per site increasing from 29.31 ± 7.65 in the 1930s to 40.18 ± 16.41 in 2009. No change in β-diversity was recorded. However, γ-diversity increased, with the total number of species rising from 219 in the 1930s to 280 in 2009. Species composition shifted over time, associated with a decline in ‘stress-tolerant’ species typical of species-rich calcareous grasslands, and an increase in species typical of mesotrophic grasslands. This was associated with an increase in mean Ellenberg N value, suggesting that eutrophication has been a driver of floristic change.

4. Elements of metacommunity structure analysis indicated that the structure of this grassland plant metacommunity was Clementsian at both survey times, indicating species sorting. Metacommunity structure was stable over time, despite changes in α- and γ-diversity. Analysis of potential structuring mechanisms revealed a significant influence of elevation.

5.Synthesis. This investigation provides a rare example of the long-term dynamics of a plant metacommunity. Results indicate that substantial change has occurred in the composition of calcareous grasslands during this time, both at local and regional scales. The investigation provides evidence of the impact of environmental change on immigration and extinction processes operating in calcareous grasslands at different scales, and highlights challenges for their future conservation.


Calcareous grasslands are recognized to be of high conservation value throughout Europe, and they support a high diversity of plant and animal species (Poschlod & Wallis De Vries 2002), many of which are rare or endangered (Webb, Drewitt & Measures 2009). Such grasslands are generally remnants of previously more extensive habitats created by traditional, low-intensity agriculture. As well as being of high cultural value (Zechmeister et al. 2003; Haines-Young et al. 2006), calcareous grasslands provide multiple ecosystem services including pollination (Carvell et al. 2006), storage of pollutants and carbon sequestration (Janssens et al. 2005). The decline in area of calcareous grasslands, caused mostly by agricultural intensification, has been one of the most dramatic habitat losses of the last century in the UK and in other parts of western Europe (Fischer & Stöcklin 1997; Eriksson, Cousins & Bruun 2002; Polus et al. 2007). For example, Fuller (1987) reported a 97% loss of semi-natural enclosed grasslands in England and Wales between 1930 and 1984.

Previous research on the plant ecology of calcareous grasslands has focused on the processes responsible for maintenance of the high species richness often observed (e.g. Mitchley & Grubb 1986), and on the dynamics of individual species at population (e.g. Schleuning & Matthies 2009) and metapopulation (e.g. Herben et al. 2006) scales. However, relatively little research has been undertaken on the long-term dynamics of calcareous grassland communities. As an example, Bennie et al. (2006) examined long-term vegetation change in British chalk grasslands by resurveying 92 plots that were first surveyed 50 years earlier. Results showed declines in species richness and in the frequency of relatively stress-tolerant species typical of species-rich calcareous grasslands. These trends were associated with a significant increase in Ellenberg fertility values, indicating increasing nutrient enrichment over recent decades.

Most previous research into the dynamics of calcareous grassland communities has focused on individual sites; much less is known about community dynamics at the scale of landscapes or regions. Available evidence suggests that in order to obtain a complete understanding of the effects of habitat loss and other drivers of change on calcareous grasslands, studies should be conducted across multiple, linked communities within landscapes. For example, Adriaens, Honnay & Hermy (2006) examined species richness and environmental characteristics in 64 calcareous grassland sites in southern Belgium. Results highlighted a positive relationship between species richness and area of grassland fragments, but no relationship with historical area or connectivity, suggesting that species may quickly be lost in response to new fragmentation events. In contrast, in a pan-European study of 147 fragmented grassland remnants, Krauss et al. (2010) found that present-day species richness of long-lived vascular plant species was better explained by past than current landscape patterns, providing evidence of an extinction debt. However, information is needed on the dynamics of community composition as well as patterns of species richness.

Understanding the ecological patterns and processes of multiple communities has been supported by the development of the metacommunity concept, which aims to unify spatial and community ecology (Leibold et al. 2004; Holyoak, Leibold & Holt 2005; Presley, Higgins & Willig 2010). Leibold et al. (2004) define a metacommunity as a set of local communities of multiple potentially interacting species that are linked by dispersal. The concept provides a useful framework for assessment of community change over broad spatial scales. Metacommunity theory, which aims to explain the distribution, abundance and interactions of organisms at the scale of multiple communities, has developed rapidly in recent years. However, empirical testing of such theory is less well advanced (Driscoll & Lindenmayer 2009; Pandit, Kolasa & Cottenie 2009). Leibold et al. (2004) identified four distinct paradigms under which theoretical and empirical work on metacommunities has developed, namely ‘patch dynamic’, ‘species sorting’, ‘mass effects’ and ‘neutral’. In the patch dynamics paradigm, spatial dynamics are dominated by local extinction and colonization of species in habitat patches, whereas the species-sorting perspective emphasizes spatial niche separation according to gradients in resources (Leibold et al. 2004). The mass-effect paradigm focuses on the role of immigration and emigration in the spatial dynamics of local population densities, whereas in the neutral perspective, all species are considered to be similar in their competitive ability, movement and fitness (Hubbell 2001). The extent to which these different paradigms apply to field situations remains unclear, and there is therefore a need to test them with empirical data (Driscoll & Lindenmayer 2009; Pandit, Kolasa & Cottenie 2009).

Few previous investigations have examined the structure and composition of meta-communities over time. Examples are provided by analyses of terrestrial gastropod metacommunities over 13 years (Bloch, Higgins & Willig 2007) and of tree-hole mosquitoes over 26 years (Ellis, Lounibos & Holyoak 2006). With respect to plants, the only previous analysis of long-term dynamics in metacommunity structure was undertaken by Keith et al. (2009), who examined change of a southern English woodland metacommunity at two points in time, seven decades apart. Results demonstrated that metacommunity structure was maintained over time, despite the fact that β-diversity declined as a result of taxonomic homogenization. The phenomenon of taxonomic homogenization refers to an increase in similarity of species composition across a set of communities (Olden & Rooney 2006). This issue has attracted increasing research interest in recent years, as it provides an indication of biotic impoverishment resulting from environmental change (Olden et al. 2004; Rooney et al. 2007). However, evidence for taxonomic homogenization is conflicting; while it has been documented in some areas as a result of invasive species (Castro, Munoz & Jaksic 2007; McKinney & La Sorte 2007; Magee, Ringold & Bollman 2008), climate change (Britton et al. 2009), eutrophication and changing management (Keith et al. 2009), it was not identified in an analysis of Mediterranean floras (Lambdon, Lloret & Hulme 2008) or in two states in the USA (Qian, McKinney & Kuhn 2008).

The analyses of Keith et al. (2009, 2011) were based on resurveys of field plots that were established in the English county of Dorset during the 1930s (Good 1937, 1948), providing a rare opportunity to examine changes in metacommunity structure and composition in native woodlands over a period of seven decades. This study presents a parallel investigation conducted over a similar interval, focusing on the calcareous grasslands of Dorset. The specific objectives were to test whether: (i) any changes had occurred in site occupation for all of the species encountered, (ii) any changes in α-, β- and γ-diversity had occurred between the two points in time, (iii) metacommunity structure differed between the two sampling points, (iv) taxonomic homogenization had occurred and also to examine (v) which factors were responsible for any differences observed. Based on the results of Keith et al. (2009, 2011) in their analysis of Dorset woodlands, we hypothesized that α- and γ-diversity and metacommunity structure would remain unchanged over time, but that β-diversity would decline, providing evidence for taxonomic homogenization.

Materials and methods

Grassland Surveys

The ‘good’ survey

From 1931 to 1936, Ronald Good undertook a survey of vascular plant species at 7575 sites throughout the southern English county of Dorset (Horsfall 1989; Wichmann et al. 2008). Good selected sites using what he referred to as the ‘stand’ method. Stands were ‘…reasonably distinct topographical and ecological entit[ies]…’ and were required to be ‘…as evenly scattered as possible’ across Dorset (Good 1937). Stands were surveyed by recording all vascular plant species encountered during a survey undertaken by traversing each individual site, with a duration of c. 1 h. Stand locations were recorded on a series of six-inch Ordnance Survey (OS) maps (Webb 1997), which were subsequently digitized by the Dorset Environmental Records Centre (DERC). Each stand was visited once, generating presence–absence data. While lacking information on the relative abundance of different taxa on each site, such data are relatively robust to sampling error (Hirst & Jackson 2007).

Resurvey of grassland sites

For clarity, henceforth we refer to Good’s stands as ‘sites’ and the species list for a site as a ‘community’. We resurveyed a selection of the sites classified as grassland by Horsfall (1989) and as calcareous grassland by DERC (C. Steele, personal communication). We selected a sample of sites for potential resurvey by excluding all sites recorded as destroyed (212 sites) or part-destroyed (64) during a field survey conducted in the 1980s (Horsfall 1979, 1980, 1981, 1984a,b, 1986). Destruction comprised conversion to other habitat types, including agriculturally improved grassland, as well as building and other construction work. Resurveys were attempted in the 174 remaining sites. Following Bennie et al. (2006), sites were selected for re-survey only if they remained in 2009 as recognizably calcareous grassland of unimproved or semi-improved character, with no evidence of ploughing, reseeding or widespread fertilizer application. Selected sites varied in size from 0.02 to 31 ha. Typical National Vegetation Classification (NVC) communities were CG2, CG4 and CG6 (Rodwell, 1992; Appendix S1 in Supporting information).

Sites were relocated in the field using a Global Positioning System (eTrex venture; Garmin Ltd., Southampton, UK) supported by digital maps of the Good sites derived from DERC and 1:25 000 scale raster OS tiles. Each site was surveyed on an ordinal date as close as possible to the date employed by Good in the 1930s and was searched over c. 2 h to maximize chances of recording all species present. This represents an increase over the survey times employed by Good, with the aim of recording all of the species present, but this may have resulted in a bias towards increased species richness values (Appendix S1). Field surveys were conducted during the summer months of 2009 by one of the authors (RMW) with extensive experience of conducting floristic surveys of calcareous grassland in the area. All vascular plant species were identified in situ. A few taxa were identified only to genus by Good; these were assumed to be the same species as specimens of the same genus found in a site in 2009 (Appendix S1). Observations of current or previous site management were recorded during each site visit including indications of cutting, grazing, agricultural improvement, conversion to arable. Sward height was measured to the nearest 5 cm using a ruler.

Data Analysis

All analyses were conducted in r version 2.9.2 (R Development Core Team, 2008). Multivariate analysis and analysis of similarity were undertaken using functions included in the ‘vegan’ package (Oksanen et al. 2008). Analysis of species traits employed values published in the PLANTATT data base (Hill, Preston & Roy 2004). Further details of the methods used are provided in Appendix S1.

Individual species

We tested the significance of site occupation change for each species using McNemar’s test (McNemar 1947), determined using the following equation with a Yates’ correction:


where b represents the number of sites at which the species became extinct between the two surveys, and c represents the number of sites that the species had colonized.


The significance of any change in species richness at the scale of individual sites (i.e. α-diversity) between surveys was tested using a paired t-test. Shapiro–Wilk tests provided no evidence that the assumption of normality was violated for both the Good survey and the 2009 survey data. Regression was used to examine the relationship between species richness in the 1930s and in 2009.

β-diversity and community composition

Owing to debate surrounding the appropriate metric for quantification of β-diversity (Anderson et al. 2011), we calculated both classical and multivariate measures. Whittaker’s classical measure divides γ-diversity by mean α-diversity for the region (β = γ/inline image), thus providing a measure of how much the regional richness is greater than richness within a smaller unit (Whittaker 1960). The more usual measure when exploring biotic homogenization is a multivariate measure (Anderson et al. 2011) that compares mean similarity in species composition among communities (Olden & Rooney 2006). Specifically, we employed the method of Anderson (2006) and Anderson, Ellingsen & McArdle (2006), which quantifies this measure based on distance from a survey centroid in multivariate space. Following this method, β-diversity for each survey was measured as the median dissimilarity of all communities within a single survey from the centroid of that survey in multivariate space, using Sørensen’s index. Dissimilarity was based on species composition within the communities; therefore, β-diversity in this case represented the variation among communities within each survey. The significance of difference in the median (M) distance from centroid between surveys was evaluated using a Wilcoxon paired test. This equates to a test of taxonomic homogenization, because in the event the difference was significant, a directional difference of M1930 < M2009 would indicate differentiation, whereas M1930 > M2009 would indicate homogenization.

The difference in species composition between the two surveys was analysed using a one-way analysis of similarity (ANOSIM) with 1000 permutations, and visualized with non-metric multidimensional scaling (NMMDS) (Everitt 1978). The ordisurf function in the r package ‘vegan’ (Oksanen et al. 2008) was applied to guide interpretation of the ordination results relating to site species richness. This function fits a smooth surface for site species richness using general additive models.

γ-diversity and the species pool

The total species richness across the total set of sites at each survey period is the simplest measure of γ-diversity, which is also referred to here as the species pool. As this is a single point estimate that uses all the observations, a bootstrap procedure was applied in order to evaluate the significance of a change in γ-diversity (Jost 2006). We drew 88 sites at random from the pooled data set and identified the total number of species in each sample. This procedure was repeated 10 000 times. This enabled z scores to be calculated for the observed species number in each of the surveys, using:


where inline image is the mean of the set of bootstrapped samples, Sobs is the observed value in the survey and SD(Sboot) is the standard deviation of the bootstrapped samples. The components of γ-diversity were compared between surveys, after dividing the overall diversity into its components. This was achieved using the following equation:


where S1930 represents the number of species unique to the first survey, S2009 represents species unique to the second survey and Sboth are species found in both surveys.

Metacommunity structure

Elements of metacommunity structure (EMS) analysis was used to determine which of the six structures best describes the metacommunity (Presley, Higgins & Willig 2010), following Keith et al. (2011) (Appendix S1). These six structures are random, chequerboard, nested, evenly spaced, Clementsian, and Gleasonian (Leibold & Mikkelson 2002), which are based on whether or not the metacommunity conforms to each of three elements, namely (i) coherence (gaps in species occurrence called embedded absences), (ii) species turnover in space (swapping of species between communities), and (iii) clumping of range boundaries. EMS was performed following optimal packing of the matrix via reciprocal averaging (RA) (Hill 1973). RA strives for maximum correspondence between species scores and samples scores with an objective order of sites and species that are maximally packed by placing the species with the most similar distributions, and sites with the most similar compositions, near to each other. The analysis included all species found in more than one site (n1930 = 177, n2009 = 200) to prevent bias when calculating coherence.

We conducted an EMS analysis to determine the structure of the grassland plant metacommunity at the two points in time, i.e. 1930s and 2009. The EMS analysis followed Leibold & Mikkelson (2002) and was undertaken in matlab (version 2008b Student) with a freely available script ( designed by Presley et al. (2009). We followed Presley et al. (2009) in conducting the analysis on both the primary and secondary axes. Changes in individual elements of coherence, spatial turnover and range boundary clumping were assessed with z scores (= (X−μ)/σ; where = observed value, μ = mean, σ = standard deviation), following Keith et al. (2011). A z score difference between the two surveys of more than twice the standard deviation (i.e. 2 × 1.96 = 3.92) was interpreted as statistically significant. Morisita’s Index (Morisita 1957) was calculated to examine the degree of overlap among samples and was applied within the EMS analysis to determine the clumping of species range boundaries.

We analysed the potential mechanisms influencing 2009 metacommunity structure with correlation tests between RA axis scores and environmental variables, which were analysed using arcgis version 9.3 (ESRI, Redlands, CA, USA). We created a point layer with a centroid for each grassland site polygon and used the arcgis intersect tool to obtain environmental data for each site. The environmental variables used were soil drainage and soil fertility (obtained from the National Soil Resources Institute 1:250 000 NATMAP data) and elevation (as a proxy for multiple climatic variables). We determined elevation for each site centroid with OS map contours, taking the lower contour if the site centroid fell between contour lines. Environmental data were only available for 2009; therefore, it was only possible to investigate potential structuring mechanisms for the resurveyed sites. The ordisurf function in the r package ‘vegan’ (Oksanen et al. 2008) was also used to investigate the relationship between mean species Ellenberg values, sourced from the PLANTATT data base (Hill, Preston & Roy 2004), and the ordination axes by using species traits as input.

Drivers of change

We assessed whether management observations recorded during the 2009 survey were correlated with the change in species richness within sites using Spearman’s rho tests (Salkind 2007). Owing to multiple tests on the same data, significance was assessed in light of a Dunn–Ŝidák correction (Salkind 2007) for experiment-wise error (αe = 1 − (1 − αr)1/k, where αe represents the new significance value, αr is the required significance value, and k is the number of independent tests.

The PLANTATT data base is a collection of attributes on status, size, life history, geography and habitats for the British and Irish plant species (Hill, Preston & Roy 2004). Selected attributes sourced from this data base were used to examine whether species that were gained or lost between the two sample dates differed in terms of their characteristics, including environmental tolerances, and national trends in their abundance. Attributes that were analysed included the Ellenberg values for N (nitrogen), L (light) and R (pH), habitat associations and plant height. PLANTATT also provides information on climatic tolerances of species. Here, we analysed mean January temperature, using the mean value across all 10-km grid squares in which the species occurs nationally, based on the UKCIP 1961–90 baseline data (Hulme & Jenkins 1998). Information on the national trends in abundance is provided as the National Change Index (NCI), which measures the relative magnitude of change in frequency at the national scale between 1930–60 and 1987–99 (Hill, Preston & Roy 2004). We examined whether mean values of these traits different between those species that were unique to either the first or second survey, and those that were found within both surveys. To determine whether it was appropriate to test for differences using parametric or nonparametric methods, we tested for heterogeneity of variances of the traits between groups of species using Levene’s tests (Levene 1960). In light of the result from these tests, differences were examined using nonparametric Kruskal–Wallis tests. Mean Ellenberg trait values were also calculated for each site based on species composition. We tested the significance of a change in Ellenberg number and NCI over time using paired t-tests, because the errors were found to be approximately normally distributed.


Scope of survey

Of the 174 sites that were considered for inclusion in the survey, 86 were found to be inaccessible or destroyed, leaving a total of 88 that were resurveyed. Those that were excluded included 44 sites that were not visited owing to difficulties of access, one that had been built on, 21 that had been converted into arable fields, 19 that had been converted to improved grassland, and one that had undergone succession to scrub.

Individual species

In the 1930s survey, Good recorded a total of 219 plant species in the sites that were resurveyed in 2009. Of these, 190 were recorded in the 2009 survey in at least one of the sites. Only 29 species were recorded in the 1930 survey but not in the 2010 survey (Appendix S2). All of these are species characteristic of open calcareous grasslands with a mix of dry and damper areas. In addition, 90 species were recorded in the 2009 survey that were not recorded by Good (Appendix S2). These are generally species of more mesotrophic communities but are associated with a range of successional states from bare ground to open woodland. Changes in site occupation between the two surveys were significant for 64 species (McNemar’s test, < 0.05), of which 54 increased and 10 decreased in the number of sites occupied (Table 1, see also Appendix S2).

Table 1.   Change in site occupation between the two surveys
  1. P-values are the result of McNemar’s tests to determine the significance of change in the number of sites occupied. Only species with significant changes are shown. 1,1 = number of sites where species were present at both surveys; 1,0 = number of sites where species present in 1930s but not in 2009; 0,1 = number of sites where species were present in 2009 but not in the 1930s; 0,0 number of sites where the species were absent at both surveys; n1930 = total number of sites occupied in the 1930s; n2009 = total number of sites occupied in 2009; Δn = change in number of sites occupied between the two surveys. National Change Index (NCI) indicates the relative magnitude of change in frequency at the national scale between 1930–1960 and 1987–1999 (values from Hill, Preston & Roy 2004). The species are grouped into ‘winners’, which demonstrated a significant increase in the number of sites occupied, and ‘losers’, which demonstrated a significant decrease.

 Urtica dioica<0.001713941846380.28
 Arrhenatherum elatius<0.00121440232561360.37
 Brachypodium sylvaticum<0.00111335323432−0.17
 Lathyrus pratensis<0.00135374384032−0.17
 Senecio jacobaea<0.0019838331747300.11
 Cirsium arvense<0.00121634272755280.47
 Dactylis glomerata<0.001463318497728−0.06
 Lolium perenne<0.0011743037214726−0.29
 Festuca rubra agg.<0.00141227184368252.96
 Odontites verna<0.00101256212524−0.46
 Trifolium repens<0.00138428184266241.31
 Crataegus monogyna<0.0011712446184123−0.76
 Heracleum sphondylium<0.001212461326230.08
 Galium mollugo0.00116829352445210.04
 Rubus fruticosus agg.<0.0011212253133421−0.29
 Rumex obtusifolius<0.001112264223210.66
 Achillea millefolium0.005221333203555200.29
 Medicago lupulina0.00717133226304919−0.43
 Agrimonia eupatoria0.001952351143218−0.89
 Fraxinus excelsior<0.00121196632118−0.73
 Phleum bertolonii<0.00120186822018 
 Euphrasia officinalis agg.<0.0011101760112817−1.61
 Ranunculus repens0.0059825461734170.55
 Dactylorhiza fuchsii0.001131866419150.33
 Potentilla anserina0.00143186372215−0.23
 Plantago major0.0018216621024140.09
 Agrostis stolonifera<0.001701368720133.66
 Carex caryophyllea<0.00100127601212−0.2
 Rhinanthus minor<0.00110127511312−0.49
 Sambucus nigra0.00442146861812−0.75
 Carex flacca0.02737516304253110.53
 Clematis vitalba0.001001177011110
 Clinopodium vulgare0.01954156492011−0.67
 Rosa canina agg.0.00331127241511 
 Anacamptis pyramidalis0.02212117431290.55
 Centaurea scabiosa0.0125110726159−0.49
 Holcus lanatus0.0495941312637291.34
 Hypericum perforatum0.0220211752119 
 Prunus spinosa0.0042097721190.4
 Stachys sylvatica0.00400979099−0.49
 Arum maculatum0.00800880088−0.28
 Convolvulus arvensis0.0082087821080.7
 Corylus avellana0.00810879198−0.54
 Hedera helix0.00810879198−0.65
 Mercurialis perennis0.00800880088−0.65
 Sonchus asper0.021019781980.78
 Ulex europaeus0.0397210699178−0.34
 Carex humilis0.01600781077−0.01
 Geranium robertianum0.01600781077−0.41
 Poa annua0.039018791870.83
 Quercus robur0.01620779297−0.6
 Arctium minus0.03110681176−0.41
 Dryopteris filix-mas0.031006820660.03
 Matricaria discoidea0.03100682066−0.49
 Trisetum flavescens<0.001153610275125−26−0.13
 Leontodon saxatilis<0.00113273454016−240.21
 Asperula cynanchica0.00224267315031−19−0.47
 Pilosella officinarum0.01562711443317−16−0.59
 Koeleria macrantha0.0159249463318−15−0.29
 Briza media0.01438217225945−14−0.75
 Cirsium acaule0.02641218186249−13−0.52
 Thymus polytrichus0.02116207453623−13−0.64
 Festuca ovina agg.0.03816208443624−12−0.15
 Plantago media0.04335187285342−11−0.79

The species that increased most markedly in the number of sites occupied (‘winners’) were (in declining order) Urtica dioica, Arrhenatherum elatius, Brachypodium sylvaticum, Lathyrus pratensis and Senecio jacobaea, each of which increased in frequency by 30 or more sites. Of the winners, 54% (i.e. 21 species) also increased at the national scale (Table 1). Species that declined most markedly (losers) were (in declining order) Trisetum flavescens, Leontodon saxatilis, Asperula cynanchica, Pilosella officinarum and Koeleria macrantha, each of which has become absent from 15 or more sites. All of the losers also declined nationally (Table 1), with one exception (L. saxatilis).


Species richness per site increased from a mean ± SD of 29.31 ± 7.65 in the 1930s to 40.18 ± 16.41 in 2009 (t = −5.638, < 0.001), with values ranging from 9 to 43 in the 1930s and 6 to 76 in 2009. This increase in mean species richness per site was negatively correlated (rs = −0.398, < 0.001) with the sites’ species richness in the 1930s (Fig. 1). Sites that lost species were more species rich originally than sites that gained species (Mann–Whitney test, U = 487.5, = 0.026).

Figure 1.

 Relationship between the species richness recorded in the 1930s and the change in species richness observed in 2009. Each point represents an individual site. Marginal box plots show medians (as the line within each box), interquartiles range (box) and total range (dashed whiskers) for each variable. The regression line (< 0.01) is shown with 95% confidence intervals.

β-diversity and community composition

Whittaker’s classical measure of β-diversity suggested little difference in β-diversity between surveys (1930s = 7.47; 2009 = 6.97). This was supported by the second multivariate method of β-diversity quantification in which a distance matrix based on Sørensen’s index was reduced to its principal coordinates and the group centroid for each survey was calculated on the principal coordinate axes (Fig. 2a). The significance of the difference in dispersal around the centroids was examined by performing an anova on the distances to group centroids. No significant difference was detected (anovaF1,174 = 0.03, = 0.84) (Figs 2b and 3), suggesting no change in β-diversity. The fact that β-diversity was maintained over time is also indicated by the similar size of ellipse observed at both survey dates (Fig. 2a).

Figure 2.

 Non-metric multidimensional scaling analysis of 1930s and 2009 survey data. (a) Site scores. Filled triangles represent the survey from the 1930s, whereas open diamonds represent the 2009 resurvey. Contour lines show a smoothed surface of mean species richness. Dashed ellipses are drawn at one standard deviation from the centroid of the distribution and solid ellipses show the 95% confidence interval for the position of the centroid. The dark arrow connects the centroids and thus shows the direction of change in composition. (b) Species scores. Contours represent the mean Ellenberg N values for species. Names of some common species are shown in order to guide interpretation. The arrow shows the direction of the compositional shift based on the centroids of the patch scores.

Figure 3.

 Box plot illustrating β-diversity values for surveys undertaken in the 1930s and in 2009, based on the distance of non-metric multidimensional scaling sample axis scores from the centroid. Differences in median and interquartile range are given for each survey. The difference between medians is not significant (anova, = 0.84).

A significant shift in species composition was observed (R = 0.135, < 0.001) over time, which is also illustrated by the lack of overlap in the 95% confidence intervals for the 1930s and 2009 centroids (Fig. 2a). The NMMDS contours (Fig. 2a) indicate that this shift was in the direction of increased species richness. In addition, the centroid shift was associated with an increase in mean Ellenberg N value (from between contours 3.5 to 4, towards contours 5 to 5.5 (Fig. 2b; see also Appendix S3).

γ-diversity and the species pool

The number of species recorded in both surveys combined was 309. The 1930s survey found 219 species (S1930 = 29 species), and the second survey 280 (S2009 = 90 species). The number of shared species (Sboth) was 190. The γ-diversity increased over time as the number of species found in the 1930s survey was lower than that for a random sample drawn from the pooled data (= −4.65, < 0.001).

Metacommunity structure

The (RA, Appendix S1) produced eigenvalues of 0.38 (Axis 1) and 0.30 (Axis 2). The EMS analysis suggested that the structure of this grassland plant metacommunity was Clementsian at both survey times on the primary axis, whereas the secondary axis was random at both survey times (Table 2). The Clementsian structure on the primary axis was demonstrated by significant coherence, spatial turnover and species range boundary clumping in comparison with the null matrices. Coherence was indicated by a lower number of embedded absences than the mean of null simulations (1930s, = 0.002; 2009, = 0.003). Spatial turnover was more than null simulations (both surveys < 0.001; Table 2), indicating that the structure was not nested. Morisita’s index was different from the expected distribution of boundaries (< 0.001 in each case, Table 2) and the direction of the value suggested clumping of species range boundaries. The random structure of the second axis was indicated by the fact that the number of embedded absences was not significantly different from the embedded absences in null matrices (> 0.1 in each case, Table 2).

Table 2.   Elements of metacommunity structure (EMS) analysis for primary and secondary axes. For details of the analysis, see text. Values in bold are statistically significant at P < 0.05
Survey dateAxisCoherenceSpecies turnoverBoundary clumpingBest fit pattern for distribution
No. of embedded absencesMeanPSDNo. of replacementsMeanPSDMorisita’s indexP
1930Primary axis702077410.002228.221 376 828763 630<0.00185 3312.14<0.001Clementsian
1930Secondary axis766877120.825200.191 084 987669 210<0.00181 3033.00<0.001Random
2008Primary axis10 75811 7400.003335.293 439 0682 299 700<0.001193 4702.33<0.001Clementsian
2008Secondary axis11 30311 6850.163273.712 258 1471 665 100<0.001153 1802.50<0.001Random

With regard to individual structural elements, a comparison of z scores indicated that there was no significant difference in embedded absences (i.e. coherence) or spatial turnover on the primary or secondary axes between the two survey periods (Table 3). Clumping increased over time on both axes, but it was not possible to determine whether this change was significant because the z score comparison was not appropriate for use with Morisita’s index.

Table 3.   Comparison of individual structural elements of the metacommunity using z scores. No significant differences were found between primary or secondary axes at either date
Survey dateAxisz score embedded absencesDifference in z scorez score species turnoverDifference in z score
1930Primary axis−3.16 7.19 
2008Primary axis−2.930.235.891.30
1930Secondary axis−0.22 5.11 
2008Secondary axis−1.401.183.871.24

Drivers of community change

Change in species richness was significantly negatively correlated with indications that the patch had been recently cut (rs = −0.297; = 0.006). Indications of grazing, conversion to arable and sward height were not correlated with species richness change (> 0.05 in each case).

The January temperature, which indicates the temperature range with which each species is associated, was significantly higher among the species recorded only in 2009 compared with values for species that were recorded in both surveys (χ2 = 14.80, < 0.001) (Fig. 4). The apparent difference in median height between these groups was not significant when evaluated by a Kruskal–Wallis test (χ2 = 4.31, = 0.12), as it was attributable to the addition of a few exceptionally tall woody species (trees) that had a disproportionate influence on the mean. Ellenberg L (light) values were significantly higher for the set of species that were recorded only in the 1930s (χ2 = 14.09, < 0.001) than in the other groups, while Ellenberg N (fertility) values were higher (χ2 = 33.96, < 0.001) for those species only recorded in 2009 (Fig. 4). These groups also differed in terms of whether or not the species are calcareous grassland specialists (i.e. whether or not the only broad habitat with which the species is associated is calcareous grassland, according to PLANTATT; Kruskal–Wallis χ2 = 10.425; = 0.005). A higher percentage of specialist species were present in both surveys (23%, Sboth) than were either lost (7%, S1930) or gained (6%, S2009).

Figure 4.

 Mean values for PLANTATT traits grouped by occurrence over the two surveys (S1930 = species unique to the first survey, S2009 = species unique to the second survey, Sboth = species that were found within both surveys) with 95% confidence intervals. (a) Ellenberg values for light tolerance (L), (b) Ellenberg values for pH tolerance (R); (c) Ellenberg values for nitrogen tolerance (N), (d) National Change Index (change in frequency at the national scale from 1930 to 1960 and 1987 to 1999 according to Hill, Preston & Roy 2004), (e) January temperature, (f) plant height.

The test of three potential structuring mechanisms for the metacommunity revealed a significant correlation of both axes with elevation (primary axis, rs = −0.223, = 0.036; secondary axis, rs = 0.216, = 0.044). This is unexpected considering that the secondary axis was random. Soil drainage and soil fertility showed no significant correlation with RA axis scores (Table 4).

Table 4.   Spearman’s correlations of reciprocal averaging axes scores for 2009 grassland communities and environmental variables: soil drainage, soil fertility, and elevation (which provides a proxy for climatic variables)
AxisSoil drainageSoil fertilityElevation
  1. rs values are correlation coefficients. Significant correlations are highlighted in bold.

Primary axis
Secondary axis


Analysis of long-term change in plant communities is commonly associated with a degree of uncertainty, resulting from potential errors in the estimation of species losses and gains owing to imprecise location of the original plots, or differences in survey effort (Bennie et al. 2006). The current study is no exception in this regard (Appendix S1). As a consequence, the results should be viewed with caution. However, the evidence suggests that substantial changes have occurred in this calcareous grassland metacommunity over the past seven decades. In those calcareous grassland sites that are still extant, no change in β-diversity was recorded, whereas α-diversity increased significantly over time, with mean species richness per site increasing by more than 37%. In addition, γ-diversity increased by c. 28%. These trends were associated in a shift in species composition, involving a decline in ‘stress-tolerant’ species typical of species-rich calcareous grasslands, such as A. cynanchica, Briza media, Cirsium acaule, Festuca ovina, K. macrantha, P.  officinarum and Thymus spp. This was accompanied by a significant increase in the frequency of more competitive species typical of mesotrophic grasslands, such as Achillea millefolium, A. elatius, Brachypodium spp., Dactylis glomerata, Holcus lanatus, Lolium perenne and Phleum bertolonii.

Eutrophication appears to be one of the main factors responsible for these floristic changes, as indicated by the increase in mean Ellenberg N values over time and results of the multivariate analysis. Increases in soil fertility in recent decades have been recorded in a number of other long-term studies of semi-natural vegetation in the UK, which have been attributed to increasing deposition of atmospheric nitrogen (Haines-Young et al. 2003; Smart et al. 2003, 2005; Bennie et al. 2006; Keith et al. 2009). Following a survey of 68 acid grasslands in Britain situated along a gradient of atmospheric N deposition, Stevens et al. (2004) concluded that long-term, chronic N deposition has significantly reduced plant species richness, with the degree of decline related linearly to the rate of inorganic N deposition. In contrast, Maskell et al. (2010) examined the impacts of N deposition on a wider range of vegetation types, using a national data set. Results again indicated significant negative impacts of N deposition on species richness in acid grassland and in heathland, but not in calcareous grassland. This is despite the fact that Ellenberg N values increased within increasing N deposition, providing evidence of eutrophication in calcareous grassland. As noted by Maskell et al. (2010), N deposition was also correlated with mean June and January temperatures for this vegetation type, hindering differentiation of N deposition from climatic effects. Similarly, Van den Berg et al. (2011) found no effect of N deposition on species richness, but reported a decrease in species diversity and evenness in calcareous grassland plots with increasing rates of N deposition, together with declines in the frequency of characteristic calcareous grassland species and the number of rare and scarce species. As suggested by Van den Berg et al. (2011), these findings suggest that the responses of calcareous grasslands to increased N deposition might differ from those of other semi-natural ecosystems such as heathland and acid grassland, perhaps as a result of current management approaches or the effects of phosphate limitation, which is a widespread feature of calcareous grasslands (Carroll et al. 2003).

Few other analyses of long-term change in calcareous grassland communities are available for comparison, and none over the interval examined here. Bennie et al. (2006) resurveyed 92 chalk grasslands distributed across four climatic regions of the UK, after an interval of c. 50 years. Some of the results obtained were consistent with those reported here, including a marked decline in the frequency of stress-tolerant species and an increase in Ellenberg N values. As a consequence, Bennie et al. (2006) concluded that eutrophication resulting from increased N deposition has had a major impact on calcareous grasslands in recent decades, resulting in a shift towards more mesotrophic grassland communities. However, in relation to species richness, the current results contrast with this previous study. First, Bennie et al. (2006) recorded a consistent decrease in mean species number per plot, whereas here we recorded a significant increase. Second, Bennie et al. (2006) recorded a net loss of three species between their two survey dates, whereas here we observed a net increase of 61 species. When comparing these two studies, it should be noted that the plot size employed by Bennie et al. (2006) was smaller (50 m2) than that employed here, which may have contributed to the differences observed.

The increases in α-diversity and γ-diversity recorded here over time contradict previous reports indicating that increased N availability has often been associated with a decline in species richness in calcareous grassland (Willems, Peet & Bik 1993; Jacquemyn, Brys & Hermy 2003). In a review of the effects of N deposition on European vegetation, Bobbink, Hornung & Roelofs (1998) noted that in calcareous grassland, enhanced growth of some ‘tall’ grasses has frequently been observed, especially of ‘stress-tolerant’ competitor species characterized by relatively high potential growth rate and N utilization. Reduction of species richness in calcareous grassland has therefore been attributed to increased rates of competitive exclusion as a result of increased above-ground production (Grime 1990). However, in a series of fertilization experiments undertaken in the Netherlands, Willems, Peet & Bik (1993) found that neither the above-ground production nor plant growth forms were sufficient to explain the changes in species richness that were observed. Rather, species richness appeared to be dependent on maintenance of resource heterogeneity at the micro-scale. Similarly, Jacquemyn, Brys & Hermy (2003) recorded a reduction in species richness following application of N to calcareous grassland communities in Belgium, but this was mitigated by grazing or cutting of the vegetation, which both maintained small-scale environmental heterogeneity and reduced competition. N-addition experiments conducted in calcareous grasslands in the UK have reported limited effects on species composition; for example, Morecroft, Sellers & Lee (1994) found no change in the relative abundances of species of vascular plants after 3 years of N addition, and Wilson, Wells & Sparks (1995) similarly found no evidence of an impact of N addition on species composition after 2 years. Although Carroll et al. (2003) reported significant declines in both overall plant cover and the abundance of some individual species over a period of 6 years following N application, these effects were slow to become apparent, and much greater responses were recorded following phosphate addition. These results support suggestions that at least in the British context, the impact of management practices such as grazing on composition of calcareous grassland communities may be much greater than that of N deposition (Wilson, Wells & Sparks 1995), and might mitigate its effects (Van den Berg et al. 2011).

In the current investigation, the increase in α-diversity and γ-diversity may therefore be attributable partly to the effects of management, a view supported by the correlation observed between change in species richness and evidence of recent cutting. Although detailed management histories are lacking for the sites studied here, like many other calcareous grasslands in the UK (Wilson, Wells & Sparks 1995; Van den Berg et al. 2011) they are still grazed or mown, contrasting with the situation in much of mainland Europe. The current results also suggest that climate change may have influenced species composition. This was indicated by the significant correlation between elevation and the RA axis scores generated during analysis of metacommunity structure. Although elevation may correlate with other variables such as soil depth and management (Van den Berg et al. 2011), those species recorded only in 2009 displayed a higher temperature requirement than those species recorded in both surveys, supporting the suggestion of Bennie et al. (2006) that climate may affect changes in the species composition of calcareous grassland. Specifically, Bennie et al. (2006) suggested that chalk grassland swards on south-facing slopes are more resistant to invasion by competitive grass species owing to the higher incidence of summer drought events. In contrast, our results suggest that projected climate change might increase such invasion, by favouring species with higher-temperature requirements. Similarly, Van den Berg et al. (2011) reported a positive relationship between species richness and maximum temperature in the sites that they studied. Both disturbance and eutrophication may accentuate climate change effects by increasing plant turnover and thus enhancing the opportunities for invasion (Grime et al. 2000).

Interpretation of the current results can usefully be informed by consideration of the processes influencing metacommunity dynamics. Change in plant species richness at the scale of a local community will be determined by the processes of local extinction and immigration, which will be influenced by dispersal between individual communities and the surrounding matrix (Stevens 2006). Variation among local communities in any factor that influences local immigration or extinction rates is predicted to influence local richness, to a degree that is dependent on the species richness of the metacommunity (Huston 1999; Stevens 2006). A similar extinction rate was recorded here to that of Bennie et al. (2006), with 29 and 36 species lost from the total species pool in the two studies, respectively. However, the two investigations differed markedly in immigration rate, with values of 90 and 33 species added to the total species pool, respectively. The relationship recorded here between colonization and Ellenberg N values suggests that the colonizing species have primarily been derived from other, more mesotrophic grassland communities. A fuller understanding of the dynamics of this metacommunity would require information on the dispersal and establishment processes of this group of species in relation to the characteristics of individual sites, which is currently lacking. However, given the importance of spatial heterogeneity for maintenance of species richness in calcareous grasslands (Rees, Grubb & Kelly 1996; Jacquemyn, Brys & Hermy 2003; Turnbull, Manley & Rees 2005; Bennie et al. 2006), it would appear that within-site spatial heterogeneity resulting from the combined effects of eutrophication and management have provided opportunities for colonization by this group of mesotrophic species, a process that has been supported by recent climate change. Further experimental analyses would be required to test this hypothesis.

These results can be viewed in relation to the four metacommunity paradigms described by Leibold et al. (2004). Specifically, the current results appear to be most consistent with the species-sorting metacommunity concept (Leibold et al. 2004), which is based on niche theory, where the presence and abundance of species in a community is determined by environmental or habitat characteristics (e.g. Gilbert & Lechowicz 2004). Evidence suggests that this may be the most common process operating in nature (Cottenie 2005); it has previously been identified in calcareous grassland in the UK (Turnbull, Manley & Rees 2005). The close association reported here between the change in metacommunity composition and environmental variables is consistent with the species-sorting paradigm, in which propagules of all species are able to reach all habitat patches, but only survive and establish in sites where environmental conditions are suitable (Leibold et al. 2004). In contrast, the lack of any increase in β-diversity in the current analysis suggests that mass effects are unlikely to predominate (Driscoll & Lindenmayer 2009). However, the role of other metacommunity paradigms cannot be discounted, as they are difficult to differentiate in practice (Driscoll & Lindenmayer 2009). A number of empirical studies of metacommunities have reported the concurrent influence of multiple metacommunity processes (Ellis, Lounibos & Holyoak 2006; Soininen et al. 2007; Ernest et al. 2008).

The current results suggest that metacommunity structure has been stable over time, despite the increases in α- and γ-diversity that were recorded. Few previous analyses of metacommunity dynamics have been conducted, but the available evidence suggests that such stability is typical (Bloch, Higgins & Willig 2007; Azeria & Kolasa 2008). The only previous analysis of metacommunity dynamics conducted for plant species is that described by Keith et al. (2009, 2011) for woodlands in Dorset, a parallel investigation to that presented here, which similarly documented changes over a 70 year interval. As in the current example, the woodland metacommunity displayed a Clementsian structure at both survey times (Keith et al. 2009). This is further evidence of the species sorting metacommunity paradigm, with community composition being influenced by groups of species responding in a similar way to structuring factors. This is indicated by the significant coherence (i.e. gaps in species range along a structuring gradient), spatial turnover (i.e. replacements) in space, and species range boundary clumping that were recorded here.

However, the current results are in contrast to those of Keith et al. (2009, 2011) in Dorset woodlands in a number of ways. While metacommunity structure remained stable in both studies, here, both α- and γ-diversity increased, whereas in the woodland communities, α-diversity remained unchanged and γ-diversity declined, with a reduction from 391 species in the 1930s to 324 species in 2008 (Keith et al. 2009). No change in β-diversity was recorded here, but a decline in β-diversity was reported by Keith et al. (2009). Together, these results suggest that metacommunity structure is insensitive to changes in either α-, β- or γ-diversity, but they also highlight contrasting responses of neighbouring metacommunities to similar environmental trends.

Taxonomic homogenization has been reported from a number of other areas, primarily as a result of species introductions (Castro, Munoz & Jaksic 2007; McKinney & La Sorte 2007; Magee, Ringold & Bollman 2008), increasing intensity of land use or urbanization (Kuhn & Klotz 2006; Dormann et al. 2007) and climate change (Britton et al. 2009). The current results provide no evidence of taxonomic homogenization in the calcareous grasslands surveyed; rather, floristic tracking (Castro & Jaksic 2008) has occurred. These results can also be usefully compared with those obtained by Smart et al. (2006), who analysed changes in plant communities that have occurred in small random sampling plots (10–200 m2) distributed throughout Great Britain since 1978. Although these authors reported no consistent change in β-diversity, results indicated that many locations have recorded an increase in α-diversity, a pattern consistent with the current investigation. Smart et al. (2006) attribute the widespread increase in α-diversity to the effects of eutrophication, intensive disturbance and abandonment; and they also suggest that increases in α-diversity are more likely to have occurred on sites that were previously low in species richness or fertility (Smart et al. 2006). These suggestions are supported by our current results.

While this investigation provides evidence of significant change in metacommunity composition and species richness over the past seven decades, these changes need to be placed in the broader context of the loss of grassland habitat (Fuller 1987). Of the 450 sites originally surveyed by Good (1939, 1948), 61% had been partially or completely destroyed by the 1980s (Horsfall 1989 and references therein). Our resurvey indicated that of the 129 sites that had not been destroyed by the 1980s and whose fates we were able to ascertain, 32% had subsequently been destroyed, with 16% converted to arable land and 14% to improved grassland. This contradicts the suggestion made by Lack (2010) that destruction of this habitat largely ceased in the 1980s. Although few of the species encountered in either survey are considered threatened at the national scale, 72% of those taxa that were lost from the species pool and 83% of those that recorded a significant decline between the two survey dates are also declining nationally (Table 1, Appendix S2). This highlights the need for strengthening efforts to conserve remaining calcareous grassland sites and to manage them effectively. The current results suggest that such management should seek to limit or reduce the opportunities for colonization by mesotrophic species. Van den Berg et al. (2011) indicate that the management approaches currently being employed in the interests of nature conservation do not prevent the adverse effects of high rates of N deposition. This raises the question of whether current management approaches for calcareous grassland are appropriate; there may be a need to manipulate grazing or cutting regimes to reduce the floristic changes that are occurring, but further research is required to identify how this might be achieved in practice. Furthermore, current suggestions that conservation planning should focus on the establishment of ecological networks (Bennie et al. 2006; Lawton et al. 2010), with the aim of reversing habitat fragmentation and increasing connectivity through ecological restoration, would be unlikely to address the causes of the floristic changes that are occurring within grasslands.


We thank the Dorset Environmental Records Centre (DERC) for supplying the Professor Good Archive records and stand location data. The National Soil Resources Institute provided soils data on licence to the Centre for Ecology and Hydrology. Thanks to Dr E. Cantarello for assistance with artwork.